Application of multiple linear regression and machine learning algorithms to elucidate the association of poor glycemic control and hyperhomocysteinemia with microalbuminuria
Microalbuminuria is an early biomarker of general vascular dysfunction and a predictor of risk for cardiovascular and renal diseases. It is also considered as a marker of insulin resistance in both diabetic and non-diabetic patients. The rationale of this study was to elucidate threshold values of fasting blood glucose (FBS) and glycosylated hemoglobin (HbA1c) that are associated with microalbuminuria. In the parallel association of microalbuminuria with hyperhomocysteinemia was investigated. Machine learning algorithm and multiple linear regression were applied to study the association of poor glycemic control on microalbuminuria and hyperhomocysteinemia. In non-diabetic subjects with FBS <102 mg/dL and HbA1c <6.3%; and in diabetic subjects with good glycemic control (FBS: 102-118 mg/dL; HbA1c: 6.3-7.0%), urinary microalbumin levels were <40µg/mg creatinine. Poor glycemic control (FBS >172 mg/dL and HbA1c >9.0%) was associated with microalbumin >40µg/mg creatinine. Age, gender, HbA1c and FBS were shown to explain variability in urinary microalbumin to the extent of 54.4% as shown by multiple linear regression model. Analysis of variance (ANOVA) revealed higher levels of FBS (F: 39.77, P <0.0001), HbA1c (F: 64.31, P <0.0001) and total plasma homocysteine (F: 3.69, P =0.04) in microalbuminuria and clinical microalbuminuria groups when compared to subjects with normal microalbumin levels. Diabetic patients with poor glycemic index had a more B12 deficiency. Poor glycemic index and hyperhomocysteinemia were associated with clinical microalbuminuria.
Diabetes; HbA1c; Homocysteine; Machine learning algorithms; Microalbuminuria
Full Text: PDF (downloaded 222 times)
- There are currently no refbacks.